This repository contains a collection of tools for working with behaviour trees using godot and C#, based on Mina Pêcheux's AI ebook that can be found here:
https://minapecheux.com/projects/ebook-unity-ai
The book takes a look at various AI models and includes example projects that demonstrate some specific concepts, like using a Behaviour Tree. I was researching the topic for a project and used this design in a game, but also ported the book's toolkit to godot (as a proof of concept for myself more than anything).
Take a look in the BehaviourTree folder as this contains the basic Node definition, state and data for use with specialised (derived) classes. The folder also has Composite and Decorator Nodes.
Selector- this acts like an 'OR', succeeding if any child node has aSuccessstate.Sequence- evaluating child nodes only succeeding when child nodes all child nodes have aSuccessstate; an 'AND'.Inverter- flips the results of a child node; e.g. 'NOT'.Timer- run a child node periodically.
Nodes form the BTree that hooks into godot's physics process callback (via Node2D) to evaluate the tree.
The Game folder contains a working example from the book that builds on the behaviour tree toolbox. It's a little RTS-like system that has a couple of AI Collectors roam around and find resources, collect them and return to base.
Each collector has an instance of the behaviour tree and each node in the tree derives from the toolbox and implements game-specific code to evaluate state, such as searching for a resource type or navigating the world. The tree is evaluated left-to-right, depth-first.
The example uses:
- tilemap with custom data (although I found this buggy using per tile data?)
- navigation agents
- sprites
- event bus
Albeit not specific to behaviour trees, the example shows how using godot's features, when broken down into discrete pieces of logic, simplifies some of the complexity associated with AI.
Do note that I don't claim this is the correct or best way to implement a game and bugs do exist (for example, finding the nearest target results in an 'as the crow flies' result; the AI should be checking the length of a navigation path result). These are not issues with the behaviour tree itself but implementation details present in all games.
The game also has some tests although no means extensive. The project uses GoDotTest.
Free AI content here: https://www.gameaipro.com/
Assets from kenney: https://kenney.nl/
Marvel's Spider-Man AI Postmortem: https://www.youtube.com/watch?v=LxWq65CZBU8
PVS-Studio - static analyzer for C, C++, C#, and Java code.